Newton Raphson Method for Load Flow:
(1)
(2)
The above equations can be rewritten in polar coordinates separating the self and
mutual admittance terms as follows
(3)
(4)
Now, = Pi,sch – [ ] (5)
= Q i,sch -[ ]
(6)
Evidently each mismatch at a bus is a function of a constant (i.e. scheduled injection)
term and the bus voltage magnitudes and phase angles. So doing Taylor’s series
expansion with respect to the unknown variables, neglecting partial derivatives of
order greater than 1 and assuming N=4, the mismatch at a bus i (nonslack) can be
written as follows
or,
Assumed, slack bus No. = 1
Similarly the reactive power mismatch equation at a bus is
(7)
If bus 4 is a
P- |V| bus this
row and column
will be excluded
•which is 2(N-1)x2(N-1) if all the buses excepting the slack bus are load bus in an N bus system.
•which is (2N-Ng-2)x(2N-Ng-2) in an N bus system if Ng number of buses are voltage controlled
with specified P i,sch and |Vi| but unknown Qi so that corresponding ΔQi (being not definable )
and Δ|Vi| (being 0) are both excluded in equation 7 above.
Expressions for partial derivatives derived from eqns. 3 and 4:
J11 elements J12 elements
J21 elements J22 elements
•Evidently Jacobian elements depend upon the |V| and δ variables and require
re-computation and inversion of the matrix in each iteration.
Algorithm for NR method:
3 and 4
5 and 6
that needs
inversion of
Jacobian matrix
•Iterate until ΔPi and ΔQi at each bus with unknown variables are less than or equal to a tolerance margin
e.g. 0.001 pu
Example 9.5: look into the worked out Jacobian and other aspects of application of basic NR
for the load flow solution for the 4 bus system which converged in only 3 iterations
e.g. Ploss in line 1-2 = P12+P21
Illustration of MW, MVAR flow with respect to line 1-3 after LF solution
Charging MVAR
Disadvantages of Basic NR method:
It is the most accurate, the fastest and with convergence almost independent of system size but too
complicated as it needs computing the Jacobian matrix and its inversion (though implicitly inverted using
triangular factorization) in each iteration.
To overcome this ,following variants have been evolved which compared to basic NR are in ascending
order of iterations required (i.e. slower convergence) while descending order of accuracy (yet
acceptable depending upon the application), time per iteration and total solution time. Their
convergence is also independent of system size.
#1) Decoupled NR: J12 and J21 i.e. off diagonal sub matrices of the Jacobian are neglected considering
weak influence of voltage magnitude changes upon the real power and the weak influence of phase
angle changes upon the reactive power so that two separated equations result as follows. Still Jacobian
diagonal blocks need to be computed and inverted in each iteration.
#2) Fast decoupled NR(though in some literature this is termed decoupled): Here based on the
characteristics of power system physical and operational parameters, a set of simplifications and
justified assumptions is applied to make the elements of Jacobian diagonal blocks (J11 and J22 ) of the
decoupled NR constant (which are the negative of the susceptance parts of the bus admittance
matrix elements) requiring evaluation and inversion only once outside the iterative loop.
Set of simplifications and justified assumptions:
angles are implied to be in radians
dividing each row of 9.86 by corresponding bus voltage magnitude on both sides
Thus both coefficient matrices J11 and J22 are made identical designated as B ̅ matrix which is generally
symmetrical and sparse with non zero elements being constant, real numbers exactly equal to negative part
of the susceptance of Ybus matrix:
•Above equations are for the case when all the buses other than the slack are P-Q buses .
For a P- |V| bus the corresponding row and column in the second equation will be excluded.
•Sometimes in the matrix B used for phase angle solution shunt elements and line resistances are
neglected, and off-nominal transformer tap is set as t=1.0 so that B becomes another modified
matrix B’.
#3) DC Load Flow: Here all voltage magnitudes are assumed to remain fixed at nominal
values of 1.0 pu so that Δ|V|=0 and only phase angle changes Δδ are determined using the
constant B’ matrix in each iteration. This is usually applied for a fast assessment of the
contingency effects.
Indeed, DC load flow can be solved in non-iterative way simply as [δ]= [B’]-1[P]